We defined categories in this experiment analogously to those in Experiment 4. The extension of each category encompassed 1/64 (1/2 x 1/2 x 1/2 x 1/8) of the representational space and thus was constrained principally on one of the four object input dimensions. Sixteen overlapping categories and 16 non-overlapping categories were defined. Four categories within each set were restricted in their range of variation principally on one of the four input dimensions. We trained separate networks to learn the overlapping and non-overlapping categories. For the overlapping categories, four linguistic context inputs specified the relevant input dimension and the subclass of outputs. In the non-overlapping case, four linguistic inputs provided redundant information about subclasses of outputs and thus were not necessary to distinguish a correct from an incorrect category.